International Journal of Engineering Research and Applications (IJERA) is an open access online peer reviewed international journal that publishes research and review articles in the fields of Computer Science, Neural Networks, Electrical Engineering, Software Engineering, Information Technology, Mechanical Engineering, Chemical Engineering, Plastic Engineering, Food Technology, Textile Engineering, Nano Technology & science, Power Electronics, Electronics & Communication Engineering, Computational mathematics, Image processing, Civil Engineering, Structural Engineering, Environmental Engineering, VLSI Testing & Low Power VLSI Design etc.
1. Sonali Mane et al Int. Journal of Engineering Research and Applications
ISSN : 2248-9622, Vol. 4, Issue 2( Version 1), February 2014, pp.557-563
RESEARCH ARTICLE
www.ijera.com
OPEN ACCESS
Image Fusion of CT/MRI using DWT , PCA Methods and Analog
DSP Processor
Sonali Mane1, S. D. Sawant2
1
Department Electronics and Telecommunication, GSM college of Engineering, Pune University, India
Professor, Department Electronics and Telecommunication, Sinhgad Technical Institute, Pune University,
India
2
ABSTRACT
Medical image fusion is a technique in which useful information from two or more recorded medical images is
integrated into a new image to offer as much details as possible for diagnosis. The fusion of different modality
images are Computer Tomography (CT) and Magnetic Resonance Imaging (MRI) by integrating the DWT &
PCA methods. The decomposed coefficients of Discrete Wavelet Transformation (DWT) are applied with the
Principal Component Analysis (PCA) to get fused image information. Choose decomposed coefficients by
fusion rule and using inverse DWT to get the fussed image of two modalities CT and MRI. The RMSE and
PSNR analysis shows better improvement on results. For the proposed fusion enhancement technique going to
implement on the processor based kit or will show the hardware support.
Keywords – Image fusion, DWT, PCA, DSP Processor
I. INTRODUCTION
Combining anatomical and functional
medical images to provide much more useful
information through image fusion has become the
focus of imaging research and processing. To get a
high-resolution image with as much details as
possible for the correct diagnosis is the main
objective of medical imaging. CT gives best
information about denser tissue and MRI provides
better information on soft tissue [2, 10, and 4]. Both
the techniques give special refined characteristics of
the organ to be imaged. So, it is looked-for that
fusion of MRI and CT images of the same organ
would result in an integrated image with detail
information [1,2,3,4].
Image fusion is a device to integrate multimodal
medical images by using image processing
techniques. Precisely it aims at the integration of
disparate and complementary data in order to
enhance the information visible in the images. It also
upturns the reliability of the interpretation
consequently leads to more accurate data and
increased efficacy. Besides, it has been stated that
fused data gives for robust operational performance
such as increased confidence, reduced ambiguity,
improved reliability and improved classification [1,
2, 3, 4 ]. Image Fusion applied in every field where
images are should be analyzed. For example, medical
image analysis, microscopic imaging, analysis of
images from satellite, remote sensing Application,
computer vision, robotics etc .
Different techniques are used for image fusion
through the evolution. On this many approaches are
www.ijera.com
used. They are of Spatial-domain like IHS, PCA,
averaging, brovey transformation, etc and other type
is Transformation-domain is like pyramid, wavelet,
curvelet transformation, etc. The disadvantage of
spatial domain approaches is that they create spatial
distortion in the fused image. Spectral distortion
becomes a negative factor while we go for further
processing such as classification problem [5,7,9].
Spatial distortion can be very well handled by
frequency domain approaches on image fusion. The
multi resolution analysis has become a very useful
tool for analyzing remote sensing images. The
discrete wavelet transform has become a very useful
tool for fusion. This paper presents method using
pixel level and used a data set of two modalities
CT/MRI images to fuse to get a relevant and
redundant information by using the DWT & PCA
[2,6,8,10] techniques. And then compare the two
methods with analytical values and qualitative
matrices like MSE, PSNR.
The DSP Processors such as ADSP-BF533/32/31
processors are enriched members of the Blackfin
processor family. While retaining their ease-of-use
and code compatibility benefits, they provide
significantly higher performance and lower power
than previous Blackfin processors. The processors are
totally code and pin-compatible, differing only with
respect to their performance and on-chip memory
[11, 12].
This paper is organized as section II explains the
DWT. In section III, PCA is dealt. Section IV
involves the proposed algorithm steps. Section V
gives Hardware support. Section VI presents
557 | P a g e
2. Sonali Mane et al Int. Journal of Engineering Research and Applications
ISSN : 2248-9622, Vol. 4, Issue 2( Version 1), February 2014, pp.557-563
experimental
conclusion.
II.
results
and
section
VII
gives
DISCRETE WAVELET TRANSFORM
(DWT)
Discrete Wavelet Transform (DWT) is a
mathematical tool for hierarchically decomposing an
image. With strong spatial support, the DWT
provides a compact representation of a signal’s
frequency component. DWT decomposes a image
into frequency sub-band at different scale from which
it can be perfectly reconstructed. The signal into high
and low frequency parts is split by the DWT. The low
frequency part contains coarse information of signal
whereas high frequency part contains information
about the edge components.
Two dimensional Discrete Wavelet Transform
implements image fusion. The resolution of an
image, which is a evaluate amount of detail
information in the image, is changed by filtering
operations of wavelet transform. And the scale is
changed by sampling. The DWT analyses the image
at different frequency bands with different
resolutions by decomposing the image into
approximation and detail coefficients (Gonzalez and
Woods, 1998).
2.1
www.ijera.com
applied DWT on LL1 & for 3Level decomposition
applied DWT on LL2 & finally get 4 sub-band of 3
level that are LL3, LH3, HH3, HL3 shown in Fig. 1.
At many different resolutions, a single image is
represented simultaneously (1x1, 2x2, 4x4, …,
2Nx2N). At every level create 4 new images of size
(2N-1)x(2N-1).
2.2
FliterBanks
Wavelet separately filters and down samples the
2-D data (image) in the vertical and horizontal
directions (separable filter bank). The input (source)
image is I (𝑥, 𝑦) filtered by low pass filter L and high
pass filter H in horizontal direction and then down
sampled by a factor of two (keeping the alternative
sample) to create the coefficient matrices IL (𝑥, 𝑦)
and I H(𝑥, 𝑦) .The coefficient matrices IL (𝑥, 𝑦) and I
H(𝑥, 𝑦) are both low pass and high pass filtered in
vertical direction and down sampled by a factor of
two to create sub bands (sub images) ILL (𝑥, 𝑦) ,
ILH(𝑥, 𝑦), IHL (𝑥, 𝑦) and IHH(𝑥, 𝑦) .
Representation of 2D-DWT Image
Fig 2. One level of 2-D image decomposition.
Fig 1. Image Decomposition using 2D DWT
Image representation using DWT, On the
previously registered images a transform is applied.
Coefficients for images are generated by This
operation. A fusion rule has to be established and
applied on these coefficients. The fused image is
obtained using inverse transform. At every level are
obtained two sets of coefficients, approximation (LL)
and detail (HL, LH and HH). First perform the DWT
in the vertical direction, followed by the DWT in the
horizontal direction. After the first level of
decomposition, there are 4 sub-bands: LL1, LH1,
HL1, and HH1. For each successive level of
decomposition, the LL sub band of the previous level
is used as the input. To perform DWT on 2 level
www.ijera.com
Fig 3. One level of 2-D image reconstruction.
The ILL (𝑥, 𝑦), comprises the average image
information corresponding to low frequency band of
multi scale decomposition. It could be measured as
smoothed and sub sampled version of the source
image I (𝑥, 𝑦). It represents the approximation of
source image. I (𝑥, 𝑦), ILH (𝑥, 𝑦), IHL (𝑥, 𝑦), and IHH
(𝑥, 𝑦), are detailed sub images which contain
directional (horizontal, vertical and diagonal)
information of the source image I (𝑥, 𝑦), due to
558 | P a g e
3. Sonali Mane et al Int. Journal of Engineering Research and Applications
ISSN : 2248-9622, Vol. 4, Issue 2( Version 1), February 2014, pp.557-563
spatial orientation. Multi-resolution could be
achieved by recursively applying the same algorithm
to low pass coefficients from the previous
decomposition.
Inverse 2-D wavelet transform is used to
reconstruct the image I (𝑥, 𝑦), from sub images ILL (𝑥,
𝑦), ILH (𝑥, 𝑦), IHL (𝑥, 𝑦), and IHH (𝑥, 𝑦) as shown in
Fig. 2. This involves column up sampling (inserting
zeros between samples) and filtering using low pass
L and high pass filter H for each sub images. The
image I (𝑥, 𝑦) would be constructed by row up
sampling and filtering with low pass filter L and high
pass filter H of the resulting image and summation of
all matrices.
2.3
Haar Wavelet Transform
Wavelets share some common properties. Each
wavelet has a unique image decompression and
reconstruction characteristics which lead to different
fusion results. They are not shift invariants and
subsequently the fusion methods using DWT lead to
unstable and flickering results. The fusion process
should not be dependent on the location of an object
in the image and fusion output should be stable and
consistent with the original input sequence for the
case of image sequences. Haar Wavelet Transform is
used to make the DWT shift invariant. Haar wavelets
are real, orthogonal and symmetric.
The Haar transform functions as a square matrix of
length N = some integral power of 2. Implementing
the discrete Haar transform consists of acting on a
matrix row-wise finding the sums and differences of
consecutive elements. If the matrix is split in half
from top to bottom, the sums are stored in one side
and the differences in the other. Next operation
occurs column-wise, splitting the image in half from
left to right. It stores the sums on one half and the
differences in the other. The process is repeated on
the smaller square, power-of-two matrix resulting in
sums of sums. The number of times this process
occurs can be thought of as the depth of the
transform.
Properties of Haar wavelet transform:
1. Haar Transform is real and orthogonal. Therefore
Hr = Hr*
(1)
Hr = Hr
(2)
Haar Transform is a very fast transform.
2. The basis vectors of the Haar matrix are sequence
ordered.
3. Haar Transform has poor energy compaction for
images.
4. Orthogonality: The original signal is split into a
low and a high frequency part and filters enabling the
splitting without duplicating information are said to
be orthogonal.
5. Linear Phase: Symmetric filters would have to be
used to obtain linear phase.
www.ijera.com
www.ijera.com
6. Compact support: The magnitude response of the
filter should be exactly zero outside the frequency
range covered by the transform. The transform is
energy invariant if this property is algebra satisfied.
7. Perfect reconstruction: If the input signal is
transformed and inversely transformed using a set of
weighted basis functions and the reproduced sample
values are identical to those of the input signal, the
transform is said to have the perfect reconstruction
property. If, in addition no information redundancy is
present in the sampled signal, the wavelet transform
is, as stated above, ortho normal.
III. PRINCIPLE COMPONENT ANALYSIS
The PCA involves a mathematical procedure
that transforms a number of correlated variables into
a number of uncorrelated variables. It computes a
compact and optimal description of the data set. The
first principal component accounts for as much of the
variance in the data as possible and each succeeding
component accounts for as much of the remaining
variance as possible. First principal component is
taken to be along the direction with the maximum
variance. The second principal component is
constrained to lie in the subspace perpendicular of the
first. This component points the direction of
maximum variance within this subspace. The third
principal component is taken in the maximum
variance direction in the subspace perpendicular to
the first two and so on. The PCA is also called as
Karhunen-Loève transform or the Hotelling
transform.The PCA does not have a fixed set of basis
vectors like FFT, DCT and wavelet etc. Its basis
vectors depend on the data set.
PCA is also a linear transformation that is easy to
be implemented for applications in which huge
amount of data is to be analyzed. PCA is widely used
in data compression and pattern matching by
expressing the data in a way to highlight the
similarities and differences without much loss of
information.
3.1
PCA Algorithm
Let the source images (to be fused) of size m x n
matrix, the steps for PCA are:
1.
Organise the data into column matrix. The
resulting matrix Z is of dimension 2 x n.
2.
Calculate the empirical mean along each
column. Assing The empirical mean matrix
M of 1 x 2.
3.
Subtract the empirical mean vector M from
each column of the data matrix Z. The
resulting matrix X is of dimension 2 x n i.e
variance.
559 | P a g e
4. Sonali Mane et al Int. Journal of Engineering Research and Applications
ISSN : 2248-9622, Vol. 4, Issue 2( Version 1), February 2014, pp.557-563
4.
Find the covariance matrix C of X
i.e.C=XXT mean of expectation = cov(X)
5.
Compute the eigenvectors V and eigenvalue
D of C and sort them by decreasing
eigenvalue. Both V and D are of dimension
2 x 2.
www.ijera.com
REG INPUT
IMAGE I1
REG INPUT
IMAGE I2
DWT(HAAR)
LL1
6.
P1 =
3.2
and
HL1
LL2
HL2
LH1
Consider the first column of V which
corresponds to larger eigenvalue to compute
P1 and P2 as:
HH1
LH2
HH2
DECOMPOSITION COEFFICIENTS
P2 =
Image Fusion by PCA
APPLY PCA to DECOMPOSITION
COEFFICIENT
The information flow diagram of PCA-based
image fusion algorithm is shown in Fig. 4. The input
images (images to be fused) I1 (𝑥, 𝑦) and I 2(𝑥, 𝑦) are
arranged in two column vectors and their empirical
means are subtracted. The resulting vector has a
dimension of n x 2, where n is length of the each
image vector. The eigenvector and eigenvalues for
this resulting vector are computed and the
eigenvectors corresponding to the larger eigenvalue
achieved. The normalized components P1 and P2
(i.e., P1 + P2 = 1) are computed from the obtained
eigenvector. The fused image is:
If (𝑥, 𝑦) = P1I1 (𝑥, 𝑦) + P2I2 (𝑥, 𝑦)
)
FUSION RULE
LL,LH,HL,HH
IDWT(Haar)
FUSED IMAGE
PARAMETER
ANALYSIS
P1I1 + P2I2
Fig 4. Image Fusion by PCA
IV.
PROPOSED METHOD
In this paper, image fusion of two modality
images i.e. CT and MRI. Consider image I1 as CT
and image I2 as MRI. The following steps are the
proposed methods, as shown in figure 4.
Fig 5. The proposed method
www.ijera.com
The images to be fused must be registered to
assure the corresponding pixels are aligned
together i.e. registered image I1 and I2.
This method is considered as the most efficient
in terms of computation time since Haar wavelet
is used. By applying DWT (haar), the reg image
I1 and I2 are decomposed i.e. {LL1, LH1, HL1,
HH1}= DWT {reg image I1} and {LL2, LH2,
HL2, HH2}= DWT {reg image I2} to obtain the
coefficients and the reg images are subjected to
3-level decomposition.
The resulting coefficients are evaluated using
PCA for both dimension reduction as well as to
obtain best coefficients for fusion. Now, using
each coefficients of reg image I1 and I2 and
applying PCA by taking respective coefficients.
560 | P a g e
5. Sonali Mane et al Int. Journal of Engineering Research and Applications
ISSN : 2248-9622, Vol. 4, Issue 2( Version 1), February 2014, pp.557-563
The fusion rule involves multiplication of each
principle component with each decimated
wavelet coefficient adding them to obtained
fused image.
The fused image is constructed by performing
the IDWT (Haar) based on the combined
transform coefficients from previous step.
V.
SDRAM
UART
Input Image
Step1. The Blackfin Evaluation Board has 128 Mbit
SDRAM interfaced in BF532 kit. To store a huge
data (pixel), this interface will be used.
Step2. The RS2329 pin serial communication is
interfaced through UART Serial Interface peripheral.
This interface is used to communicate kit with the
Matlab.
HARDWARE SUPPORT
Analog Devices Blackfin® processors offer
best-in-class performance for the given power and
cost, allowing developers to create intelligently aware
systems that communicate via wireless or wired
connections and are not limited to a specific package
or standard. A set of image processing primitives
designed to enable faster development of complex
image or video processing solutions on Blackfin is
called Blackfin Image Processing Toolbox. Primitive
functions have been highly optimized to run on
Analog Devices’ Blackfin BF-5xx and BF-60x
processor family. It is a self-contained software
module.
All the ease of use and architectural attributes of
the Blackfin processor are offered by Analog Devices
initial product family, the ADSP-BF531, ADSPBF532, and ADSP-BF533. These three processors
are all completely pin compatible - differing solely
with respect to their performance and on-chip
memory. Thus reduces risk and offering the ability to
scale up or down depending upon the end application
needs. All three processors offer low power
consumption with scaleable performance from lowcost to very high performance.
The ADSP-BF532 offers excellent performance,
large on-chip memories, and an array of applicationtuned peripherals. The Blackfin Evaluation Board is
specially devised for developers in DSP field as well
as beginners. The BF532 kit is devised in such way
that all the possible features of the DSP will be easily
used by everyone. The kit supports in
VisualDsp++5.0 and later.
The followings are playing a major role in our
hardware support:
BF532 KIT with 128Mbit SDRAM, 1Mbyte
FLASH & UART
Visual Dsp++
MATLAB
PC
www.ijera.com
Blackfin
Step4. The MatLab will help to see images on GUI
Window from processor UART through pc.
VI.
PERFORMANCE EVALUATION
The two modality CT and MRI images are
fused by proposed method i.e integrating of DWT
and PCA methods. By using platform of MatLab, the
implementation of this project is done The GUI
model of this project implementation is shown below
in fig 7. The fused image is analyzed with MSE,
PSNR, Entropy, the values of these over proposed
method gives better results.
Fig. 7 The GUI implementation of project
In the MRI image, the inner contour is missing
but it provides better information on soft tissue. In
the CT image, it provides the best information on
denser tissue with less distortion, but it misses the
soft tissue information.
PC
Flash
Step3. The Visual Dsp++ will help to do the source
code for Blackfin 532 to implement the Image Fusion
algorithm and to debug.
Output Image
Fig. 6 The flow diagram of proposed system
www.ijera.com
561 | P a g e
6. Sonali Mane et al Int. Journal of Engineering Research and Applications
ISSN : 2248-9622, Vol. 4, Issue 2( Version 1), February 2014, pp.557-563
Fig 8. Input Image1 and input Image2 represents the
CT and MRI images of Brain of same person
respectively
Fig 9. Fused images by (a) Normal Min (b) Biorthogonal (c) PCA method (d) DWT(Haar) (e) 3level of decomposition (f) the proposed method
DWT-PCA and Analog BF532 support image fusion.
The results are compared and given in table of
graphs and respective graphs of quantitative matrices
are given. The following table shows the statistical
parameters of reconstructed images.
Table1. Statistical Parameters
Parameter
RMES
PSNR
75.71
fusiom
Methode
Normal Min
Bi-orthogonal
PCA
DWT(Haar)
DWT+PCA
(a)
www.ijera.com
29.33
76.80
75.59
25.86
13.70
29.31
29.34
34.00
36.76
(b)
Figure 10. Fusion Method Vs RMES , PSNR
(c)
(d)
The statistical parameters are Mean Square Error,
Signal to Noise Ratio and Entropy. The table shows
that the proposed technique outperforms the other
fusion method. From the above table, it can observe
that the proposed method has less Mean Square Error
and high Signal to Noise Ratio compared to other
methods.
VII.
(e)
(f)
(g)
www.ijera.com
CONCLUSION
The fusion of CT and MRI of two modality
images improves the view of the images and adds
information of both anatomical and physiological
information in one image. The Wavelet transforms is
the best technique for the image fusion which
provides a high quality spectral content. But a good
fused image has both quality so the combination of
DWT & spatial domain fusion method (like PCA)
fusion algorithm improves the performance as
compared to use of individual DWT and PCA
algorithm. Finally this review concludes that a image
fusion algorithm based on combination of DWT and
PCA with morphological processing will improve the
image fusion quality and may be the future trend of
research regarding image fusion.
562 | P a g e
7. Sonali Mane et al Int. Journal of Engineering Research and Applications
ISSN : 2248-9622, Vol. 4, Issue 2( Version 1), February 2014, pp.557-563
www.ijera.com
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
Soma sekhar. A, Giri Prasad M. N,’Novel
approach of image fusion on MR and CT
images using wavelet transforms’ IEEE (2011).
Deepali Sale, Dr. Madhuri Joshi, Varsha Patil,
Pallavi Sonare and Chaya Jadhav, Image
Fusion For Medical Image Retrieval,
International Journal of Computational
Engineering Research Vol, 03 Issue, 8,2013.
Shih-Gu Huang „wavelet for image fusion.
Aditi Rana and Shaveta Arora, Comparative
Analysis
of
Medical
Image
Fusion,
International Journal of Computer Application
(0975-8887) Volume 73- No.9, 2013.
V.P.S. Naidu and J.R. Raol, Pixel-level Image
Fusion Using Wavelet and Principal
Component Analysis, Defence Science Journal,
Vol. 58, No. 3, May 2008, pp. 338-352.
Yong Yang, Multiresolution Image Fusion
Based on Wavelet Transform By using a novel
technique for Selection Coefficients,
S. Vasuki, S. Gandhimat, S. Manicka
Vinodhini, Comparative Analysis of Wavelets
for
Fusion
Application,
International
Conference on Emerging Technology Trends on
Advanced Engineering Research(CETT’12).
Mohamed I. Mahmoud, Moawad I. M.
Dessouky, Salah Deyab, and Fatma H.
Elfouly, Comparison between Haar and
Daubechies wavelet Transformions on FPGA
technology, PROCEEDINGS OF WORLD
ACADEMY OF SCIENCE, ENGINEERING
AND TECHNOLOGY VOLUME 20 APRIL
2007 ISSN 1307-6884.
[9] Deepak Kumar Sahu, M.P.Parsai, Different
Image Fusion Techniques – A Critical Review,
International Journal of Modern Engineering
Research (IJMER) Vol. 2, Issue. 5, Sep.-Oct.
2012 pp-4298-4301 ISSN: 2249-6645.
[10] Kusum Rani, Reecha Sharma, Study of
Different
Image
Fusion
Algorithm,
International Journal of Emerging Technology
and Advanced Engineering (ISSN 2250-2459,
ISO 9001:2008 Certified Journal, Volume 3,
Issue 5, May 2013).
[11] Getting Started With Blackfin® Processors
Revision 5.0, April 2010 Part Number 82000850-01.
[12] ADSP-BF52x Blackfin Processor Hardware
Reference (Volume 1 of 2) Preliminary
Revision 0.31 (Preliminary), May 2008, Part
Number
82-000525-03.
www.analog.com/blackfin
www.ijera.com
563 | P a g e